Let's face it - every enterprise software vendor is going to make some type of splashy generative AI announcement this year. The challenge for customers will be to determine:
Which vendor is truly innovating in AI, and which is better at writing press releases? What's ready for prime time, and what's not? What are the risks?
But there was one AI announcement I was really waiting for: Zoho. That wait is already over. Today, at Zoholics Austin, Zoho announced their generative AI plans (Zoho Integrates OpenAI with Zia, Strengthening Generative AI Capabilities While Upholding Core Tenets of AI Strategy).
So why was I waiting for Zoho? Because generative AI, for all its potential, also raises concerning issues with data privacy, bias, and governance. I've written on Zoho's stance on data privacy before - when it comes to customer data privacy, Zoho is resolute: Zoho takes their B2B data privacy stand - "We will not be part of this industry practice."
Zoho's determination to protect their customers' data privacy has led them to enterprise extremes: they've even built out their own data centers; none of their customer data resides in any of the big tech public clouds. Zoho obsesses about the "leakage" of customer data to the point that they've even built their own, newly-announced "privacy centric" browser Ulaa (which is publicly available to all, not just Zoho customers). Up to this point, Zoho has built all their own AI tools.
But hold up: how can Zoho reconcile their data privacy stance with their use of OpenAI's third party tooling? I'll give you Zoho's explanation shortly, but my take is this: Zoho felt strongly enough about the power of generative AI to get that functionality in front of customers now. However, when you dig deeper into Zoho's plans, you can see how their long term "data privacy first" position factors in.
Zoho's generative AI news - application scenarios
As for integrating generative AI into applications, Zoho's AI announcements list a slew of specific application functions: "Generative AI is the latest step along the company's innovation roadmap, blending third-party intelligence with Zia, Zoho's powerful AI engine, which runs on Zoho's secure cloud." Here is a sampling:
Zoho Analytics with Generative AI:
- Get suggestions and import public datasets into Zoho Analytics. Blend public data with business data to gather insights
- Define formulas for KPI metrics
- Create SQL queries from your questions in natural language
- Create a set of synonyms for table column names
Zoho Desk with Generative AI:
- Automatically summarize incoming and outgoing tickets
- Analyze mood of customer based on tone of request
- Generate automatic replies from knowledge base articles
- If not enough KB pieces available, track down publicly available information for solutions
Zoho Writer with Generative AI:
- Suggest headlines, titles, and better word replacements
- Fix punctuation and shorten content, when necessary
- Ask questions within Writer and integrate answers into document
This granular detail indicates Zoho's AI approach: embed AI into existing applications, with an emphasis on productivity gains, automations, and better decision-making.
Behind Zoho's generative AI plans - data privacy implications
But when it comes to generative AI specifically, Zoho has a three phase plan:
Phase 1: Leverage (third party) OpenAI, with a customer opt-in for accessing OpenAI functions, via extensions on the Zoho Marketplace.
Phase 2: Transition to open source generative AI tooling, bringing all customer data back in-house in Zoho's own data centers.
Phase 3: Build and use Zoho's own large language models and generative AI tools.
Last week, at Zoholics Jersey City, Zoho Chief Evangelist Raju Vegesna explained the plan:
We are taking a phased approach to this. First, whatever technology that we do, it is included by default within Zoho, to all our users, protecting the privacy. Now, we are doing some integration to ChatGPT across various applications. Those are not included by default within Zoho; they are available as extensions in our marketplace.
This is not an opt-out setup, it's opt-in. Vegesna:
If a customer wants it, they can install those extensions. As soon as they install the extension, they are shown a clear message saying, 'You're choosing to send your information to ChatGPT. Are you sure you really want to do it?' In plain language, making sure that the user is completely aware of what they are willing to do.
That is the reason for not enabling it by default within the Zoho tools. It is being an extension because of the privacy implications. Short-term, we are integrating with third parties like ChatGPT. Mid-term, the plan is to bring these technologies within our data center, so that the information about the customer does not leak to any third party.
Zoho is already having talks about the mid-term - customizing open source technology. But for the long-term, Zoho will, as usual, build it themselves:
Long-term, obviously, we are going to own the technology, and make sure that everything is built into the product, available to everyone, respecting the privacy and respecting the access and authorizations, all of them. Particularly in the enterprise, that is the critical thing, just because you're scanning everything does not mean that everyone has access to that information. You've got to own that access control, and then apply the model based on that.
AI and data privacy - surveillance tech, or productivity enhancers?
When I look back on Zoho's outspoken stance on B2B privacy - "We are not going to let surveillance companies track users on our properties" - I can't help but think that generative AI is bringing these issues to a head. The lure of powerful AI is going to make a lot of companies (and individuals) glide over privacy concerns. But that's not necessarily a good tradeoff. As Vegesna said to me:
Obviously, privacy has become increasingly important given AI's [advances]. During this phase, if anything, we have to raise the bar on what it should be. We're taking the stance that customer information should be restricted to the customer. The implementation, the technology that we have developed in-house, was essentially to ensure that.
There are two or three things we focus on when it comes to AI. One, it has to be useful for the customer, - and you have to do that by respecting their privacy. And: it has to be available for everyone. If you look at all the things we have done in AI, everything was free; everything was included; and everything was available to everyone. We were able to do that because we own the technology.
So how does that apply to this new generative AI technology? For now, Vegesna boils it down to several key use cases, including generating content and images, summarization, and, as we discussed later, generating code. But as Vegesna points out, generative AI is hardly the only form of AI out there - Zoho's AI development has involved somewhere around 100 different AI approaches. Vegesna does not want these latest announcements - and the generative AI fervor - to obscure the broader AI pursuit Zoho has cultivated over a decade. "These are just additional tools we're adding to our toolkit," he says.
The problem of AI bias - how does Zoho address it?
But these new tools come with the classic AI perils - not the least of which is bias. So how does Zoho protect against that? Probably the highlight of the generative AI conversations this spring is not the first use cases, but the "what's possible" discussion. Along those lines, we talked about the possibilities of putting an interactive chatbot interface in front of the software. Take HR for example. Vegesna brought up a scenario where you could ask your Zoho HR interface, "Tell me which employees have good knowledge of the construction industry?"
The possibilities of a chat-based, prompt-like interface in front of enterprise software is intriguing - especially given the historical problems with building great enterprise UIs, not to mention enterprise search shortfalls. But my concern, as I told Vegesna: you're now putting a lot of trust into that chat interface. In his construction example, if that chat interface overlooks some qualified candidates, you have a serious bias problem. What do we do about that?
Vegesna says it starts with the quality of the training data. One advantage to enterprise AI: you can potentially train that model on more controlled, company-specific data sets:
How you trained that system matters. The additional thing we are looking at is: why should it be a common training? Can the training aspects of it be exposed to each company? The training in real estate is going to be different from the construction industry, and is going to be different from other industries. The training in one country and the parameters are going to be different.
So providing some of the flexibility to the individual company, letting them train it, tune it for their needs, if it's not meeting their needs - is something that we're looking at. But even to do that, we need to have control of the technology; we have to bring it in-house. That is one of the reasons we are working on that in Austin.
It's not easy to take stock of generative AI news. Every vendor is piling on, but we're not far enough along to get the customer proof points and use case validations I typically insist on before writing about emerging tech. But I think we all know generative AI is a different type of phenomenon, and we must press on. I did find that Zoho's news came with more specific scenarios than we usually see. That points to both the domain and tech expertise, but we'll see how customers respond.
I know from attending the open Q&A at Zoholics Jersey City last week: Zoho's customers are counting on Zoho to build AI functionality that don't have the resources to build internally. They trust Zoho to do this while retaining that commitment to data privacy, but they do want to hear more about how that is going to play out. This announcement is a start; I'm sure Zoholics Austin will advance those conversations. Brian Sommer, diginomica contributor and thorny question specialist, is on the ground and should have a meaty report for us when all's said and done.
It was fascinating to talk to Vegesna about some of the ChatGPT experiments he has put to use inside and outside of Zoho. The brainstorming capabilities stand out - ChatGPT is pretty good at generating a list of potential corporate slogans, especially within parameters like "limit the slogan to exactly four words" or some such:
I asked ChatGPT to suggest a good billboard message for Zoho. I gave it a budget for 4 words. It aptly said ‘Simplify work. Amplify success’. This message is now live on a billboard in Dallas. pic.twitter.com/TY56SEg9xE
— Raju Vegesna (@rajuv) May 4, 2023
ChatGPT also excels at quickly compiling data points, such as comparing the top three agricultural crops across 20 countries - something that would take longer via traditional search.
But Vegesna notes that such informal research, while powerful, would need to be validated carefully before being used in any form of publication - or crucial public decision. Then there is developer productivity: count Vegesna amongst those who sees generative AI as leading to a major productivity jump - and perhaps change - in how develop teams are structured. I continue to insist that ChatGPT is an evolutionary technology in the enterprise. The developer/coding possibilities of ChatGPT are the one exception I see so far. This looks like more than an evolution on the coding side; time will tell.
As always, we must be careful not to let the AI news obscure everything else: Zoho has issued several other notable news items today as well, including their upmarket momentum update. I'm sure Sommer will have something to say about that.